[AI Image Generation 2] The Shock of Stable Diffusion! How It Changed the World
- Stable Diffusion ignited a rapid AI image revolution
- Numerous derivative apps have been developed
- Stable Diffusion 3.5 and Flux.1 are currently competing to become the new standard
Introduction
Hello, this is Easygoing.
The top image was generated using the custom model blue_pencil-flux1_v001 from Flux.1, which creates highly realistic anime-style illustrations.
Following up on the previous article, today we’ll continue exploring the history of AI image generation.
Stable Diffusion: The Rising Star of AI Image Generation
Today, we’ll focus on Stable Diffusion and the various apps it inspired.
gantt
title Stable Diffusion and derivative apps
dateFormat YYYY-MM-DD
tickInterval 6month
section Stable <br>Diffusion 1
Stable Diffusion 1.1 :a1, 2022-08-22,2022-08-28
Stable Diffusion 1.2 :a2, after a1, 2022-09-03
Stable Diffusion 1.3 :a3, after a2, 2022-09-13
Stable Diffusion 1.4 :a4, after a3, 2022-10-11
Stable Diffusion 1.5 :a5, after a4, 2024-10-31
section Stable Diffusion 2
Stable Diffusion 2.0 :b1, 2022-11-24, 2022-12-07
Stable Diffusion 2.1 :b2, after b1, 2024-10-31
section Stable Diffusion XL
Stable Diffusion XL 0.9 :c1, 2023-04-12, 2023-07-26
Stable Diffusion XL 1.0 :c2, after c1, 2024-10-31
section Stable Diffusion 3
Stable Diffusion 3 :d1, 2024-06-12, 2024-10-22
Stable Diffusion 3.5 :d2, 2024-10-22, 2024-10-31
section Novel AI
Novel AI Diffsuion :crit,e1, 2022-10-03, 2023-04-12
Novel AI Diffusion Anime V2 :crit,e2, after e1, 2023-11-06
Novel Ai Diffusion Anime V3 :crit,e3, after e2, 2024-10-31
section Leonardo AI
Leonardo AI :done,f1, 2023-09-20, 2024-10-31
section Kolors
Kolors :g1, 2024-07-06, 2024-10-31
section Flux
Flux.1 :h1, 2024-08-01, 2024-10-31
Flux.1.1 :h2, 2024-10-03, 2024-10-31
The Stable Diffusion Series
- Shocked the world by making its source code open
- Sparked numerous spin-off apps
- Older versions like SD 1.5 and SDXL are still actively used due to compatibility issues
Stable Diffusion debuted in August 2022, stunning the world by releasing its complete source code.
The idea of open-sourcing AI to make it a shared asset for humanity has existed before, with organizations like OpenAI originally founded on similar principles.
While many AI technologies have been open-sourced in the past to accelerate research, image generation AI’s potential societal impact meant its release was hotly debated.
From the beginning, Stable Diffusion was committed to open access, finally releasing publicly in August 2022.
Though not the very first image generation AI, its release pressured platforms like Midjourney and DALL-E to go public as well.
Stable Diffusion 1: The Game-Changer
Stable Diffusion 1 was the first publicly available model in the series. Its release catalyzed a massive expansion in the image generation community, attracting everyone from startups to individual developers.
Numerous custom models emerged, sparking immense excitement. Stable Diffusion became the largest AI image generation community, and groundbreaking tools like ControlNet soon followed.
Stable Diffusion 2: Potential but Limited Adoption
Just three months after the release of Stable Diffusion 1, Stable Diffusion 2 was launched, featuring an increased training resolution (512x512 to 768x768) and improved quality and stability.
However, Stable Diffusion 2 saw limited adoption, largely due to its incompatibility with models from SD 1.5, preventing users from utilizing earlier resources.
Stable Diffusion XL: Gradual Adoption with Time
Stable Diffusion XL was released about six months after SD2, with a training resolution of 1024x1024 and dual text encoders to enhance prompt accuracy.
Initially, SDXL faced the same adoption challenges as SD2.
However, its superior baseline performance and improvements in model efficiency gradually led to wider adoption, establishing it as one of the most popular communities in AI image generation today.
Stable Diffusion 3: Promising but Controversial
Stable Diffusion 3, launched in June 2024, marked a major update with a new MMDiT architecture, significantly boosting potential.
However, SD3 was criticized for its limitations in anatomical accuracy and its restrictive licensing, which initially limited widespread use.
Following a license revision in October 2024, Stable Diffusion 3.5 was released with a fully open license, rekindling interest.
Applications Inspired by Stable Diffusion
gantt
title Stable Diffusion and derivative apps
dateFormat YYYY-MM-DD
tickInterval 6month
section Stable <br>Diffusion 1
Stable Diffusion 1.1 :a1, 2022-08-22,2022-08-28
Stable Diffusion 1.2 :a2, after a1, 2022-09-03
Stable Diffusion 1.3 :a3, after a2, 2022-09-13
Stable Diffusion 1.4 :a4, after a3, 2022-10-11
Stable Diffusion 1.5 :a5, after a4, 2024-10-31
section Stable Diffusion 2
Stable Diffusion 2.0 :b1, 2022-11-24, 2022-12-07
Stable Diffusion 2.1 :b2, after b1, 2024-10-31
section Stable Diffusion XL
Stable Diffusion XL 0.9 :c1, 2023-04-12, 2023-07-26
Stable Diffusion XL 1.0 :c2, after c1, 2024-10-31
section Stable Diffusion 3
Stable Diffusion 3 :d1, 2024-06-12, 2024-10-22
Stable Diffusion 3.5 :d2, 2024-10-22, 2024-10-31
section Novel AI
Novel AI Diffsuion :crit,e1, 2022-10-03, 2023-04-12
Novel AI Diffusion Anime V2 :crit,e2, after e1, 2023-11-06
Novel Ai Diffusion Anime V3 :crit,e3, after e2, 2024-10-31
section Leonardo AI
Leonardo AI :done,f1, 2023-09-20, 2024-10-31
section Kolors
Kolors :g1, 2024-07-06, 2024-10-31
section Flux
Flux.1 :h1, 2024-08-01, 2024-10-31
Flux.1.1 :h2, 2024-10-03, 2024-10-31
NovelAI: Ideal for Anime
- Launched in October 2022 as an illustration feature in a novel-writing support app
- Enhanced from Stable Diffusion with additional anime training
- Specializes in anime-style illustrations
NovelAI Diffusion was introduced as an illustration tool within a novel-writing app but has since become a widely used image generation tool.
Released on October 3, 2022, NovelAI Diffusion quickly made headlines.
Just five days later, it suffered a hack that led to a data leak, widely circulating its resources within the Stable Diffusion community.
Certain SD1.5 custom models contain data from this breach, but SDXL is unaffected.
Anime quality has continued to improve with NovelAI Diffusion Anime V3, which is now built on SDXL.
Leonardo AI: Integrates with Canva and Supports Stable Diffusion Features
- Developed by Leonardo.Ai, built on Stable Diffusion XL
- Offers some SD features like ControlNet
- Acquired by Canva on July 30, 2024, and integrated into the platform
Launched by Australian company Leonardo.Ai in September 2023, Leonardo AI offers multiple models, including realistic and anime options.
Users can access selected Stable Diffusion features, such as ControlNet.
Following its acquisition by Canva on July 30, 2024, Leonardo AI has become available within the Canva platform.
Kolors: High-Resolution and Multilingual
- Developed by China’s Kuaishou Technology, based on SDXL
- Supports both English and Chinese prompts
- Claims higher resolution than Stable Diffusion 3
Released on July 6, 2024, Kolors was developed by Kuaishou Technology in China.
It improves prompt accuracy using the ChatGLM3-base language model, which supports both English and Chinese.
While Kolors is open source, its adoption depends on the development of its community.
Flux.1: The New Front-Runner?
- Created by former Stable Diffusion developers
- Open-source release in August 2024
- Emerging as the de facto successor to Stable Diffusion 3
Developed by a group of former Stability AI engineers, Flux.1 debuted on August 1, 2024, under the newly established Black Forest Labs.
Flux.1 features three versions:
- FLUX.1[pro]: private
- FLUX.1[dev]: open-source but not for commercial use
- FLUX.1[schnell]: lightweight, fast, open-source, and available for commercial use
With high-quality output, even in its lightweight FLUX.1[schnell] version, Flux.1 has rapidly gained popularity, especially with its integration into the X platform’s Grok assistant.
Recent advancements in quantum-quantized lightweight models have further boosted development, making Flux.1 a strong competitor to Stable Diffusion 3.
Anifusion: Create Manga in 5 Minutes
- Manga creation app based on Stable Diffusion
- Generates manga in as little as 5 minutes
- Paid service (20 Euros/month) with Flux.1 support
Anifusion is a web-based app for creating manga, allowing automatic generation from story to artwork. Users can intuitively adjust frames and speech bubbles.
I found it remarkably easy to create comics even as a beginner.
Anifusion also offers exclusive licensing for commercial manga creation using high-quality Flux.1[dev] images, providing a potential edge over future services.
Reference: Introduction to Anifusion
Summary
- Stable Diffusion ignited a rapid AI image revolution
- Numerous derivative apps have been developed
- Stable Diffusion 3.5 and Flux.1 are currently competing to become the new standard
This research has helped me organize my understanding of AI image generation, revealing significant advancements in just two years between Stable Diffusion 1 and Flux.1.
Although I’m mainly a user, I am deeply grateful to the creators behind these models.
In the next article, I’ll introduce even newer developments in AI image generation. Stay tuned!
Creator Introductions
For this post, I collaborated with two creators who provided works created with Leonardo AI and Kolors, platforms I haven’t personally used yet.
Leonardo AI: K.Kameno
生きていることは、川の水のように常に流れ、変化していくことに近い。言葉を伴う意識は、この川の水ほど自在に速やかに変化することができない。だから、自分について意識しすぎるとうまくいかないことが多い。
— K. Kameno／亀野 功／「亀の甲より年の功」になれない初老 (@K_Kameno) May 8, 2023
Photo by：https://t.co/vmvPleoOgn pic.twitter.com/J57vTaUNAS
Kolors: Browncat AI
おはようございます♪
— Browncat AI (@browncatro1) February 12, 2025
明日はバレンタインデーですが、それはともかく、今日一日を大切に過ごしましょう🧁 pic.twitter.com/4swOvtIhvE
Both creators have their own unique worlds that they express beautifully in their work.
Thank you both for sharing your creations!
Note: The copyright of the illustrations in this article belongs to the respective creators. Unauthorized reproduction is strictly prohibited.